A Comprehensive Checkpoint Mechanism For Virtual Disk Image

Recent development of virtualization technology enables us to implement ofcheckpoint/restart at a virtual machine (VM) level to provide fault-tolerance support for aVM. By establishing the checkpoint/restart mechanism on a single VM, we can improvethe VM’s reliability, and reduce operational losses. Meanwhile, it is also the foundation ofconstruction of other fault-tolerant software, and is beneficial for establishing the wholefault-tolerant system for the parallel cluster environment. A single VM’s full-statecheckpoint should include the state of VCPU register, memory information and diskinformation, etc. However, the current Xen’s checkpoints do not address the state ofexternal storage. To guarantee a VM’s persistent state to be consistent with its transientstate, the disk state correlated to the running VM should also be taken checkpoint.To save the full state of the VM, we propose a combination of memory checkpointwith disk snapshot to back up a full-state VM. Disk checkpoint is implemented byupdating the reference count and saving the current L1table in the disk, to ensure that thedisk blocks after taking snapshot will not be changed unless the snapshot is deleted.Besides, To reduce the time overhead of disk snapshot and improve the write performanceafter taking a snapshot, we propose cluster reference count storage refinement, clusterallocation and recovery management, and meanwhile sector-granularity Copy-On-Writeafter snapshot.Our system has been implemented based on the Xen virtualization technology. Thekernel modules are implemented in C language. The function test indicates that it canactually save and restore the system state for fault-tolerance. The performance test showsthat it has reduced the time overhead of saving the disk image’s state. Besides, the diskspace overhead has been reduced.